.png)
Personalization is what sets a cold email apart from spam. But manually personalizing every message is impossible on a large scale. AI has solved this problem. Here's how.
The 3 Levels of Customization
Level 1: Basic variables. First name, company name, industry, job title. That’s the bare minimum. By 2026, that won’t be enough to set yourself apart.
Level 2: Specific context . Recent LinkedIn posts, company news, ongoing recruitment, fundraising, or a change in job title. This level requires data enrichment to be performed beforehand.
Level 3: AI-generated icebreaker. A unique opening line for each prospect, automatically generated from enriched data and a specific prompt. This is the level that really makes a difference.
The Clay Workflow for Large-Scale Personalization
Clay is the central tool in this workflow. It connects to dozens of data sources (LinkedIn, Apollo, Clearbit, BuiltWith, websites) and allows you to call an AI model directly from a column in the table.
Step 1: Import the list of prospects into Clay. Add a "LinkedIn URL" column for each contact.
Step 2: Set up a "Latest LinkedIn Post" column by connecting Clay's LinkedIn API. This column automatically retrieves the prospect's latest public post.
Step 3: Create a column titled "Icebreaker AI." In this column, call Claude or GPT-4 using the following prompt:
"You're an expert in B2B prospecting. Write a personalized opening line—no more than one sentence—for a prospecting email addressed to [First Name] [Last Name], [Title] at [Company]. Their latest LinkedIn post: [Latest Post]. Your goal: to show that you've read it without repeating their exact words. Your tone: direct, professional, and free of flattery."
Step 4: Export the list—including the "icebreaker" column—to Lemlist or LaGrowthMachine. Insert the "icebreaker" variable at the beginning of the message.
Rules for an icebreaker that feels genuine
The icebreaker shouldn't sound like a generated variable. It should sound like a natural observation. Test 20 icebreakers manually before scaling up. If the sentence sounds awkward or too generic, the prompt needs to be reworked.
Avoid icebreakers that start with "I read your post about..." or "Your LinkedIn post made me..." These phrases are now recognized as automated personalization patterns.
Choose opening lines that create tension: "What you described about [topic] is exactly what we're hearing in [Industry] right now."
How this affects the results
Campaigns using Level 3 AI icebreakers generate, on average, 2 to 3 times more responses than campaigns using Level 1 personalization. For a list of 500 prospects, this translates to 15 to 30 additional responses per campaign.
To understand how to incorporate this personalization into a complete email sequence, the email sequence structure explains how to position the icebreaker in each message and how to vary the angles in follow-up messages.
